The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/vm/uma_int.h

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2002, 2003, 2004, 2005 Jeffrey Roberson <jeff@FreeBSD.org>
    3  * Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org>
    4  * All rights reserved.
    5  *
    6  * Redistribution and use in source and binary forms, with or without
    7  * modification, are permitted provided that the following conditions
    8  * are met:
    9  * 1. Redistributions of source code must retain the above copyright
   10  *    notice unmodified, this list of conditions, and the following
   11  *    disclaimer.
   12  * 2. Redistributions in binary form must reproduce the above copyright
   13  *    notice, this list of conditions and the following disclaimer in the
   14  *    documentation and/or other materials provided with the distribution.
   15  *
   16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
   17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
   18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
   19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
   20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
   21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
   22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
   23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
   24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
   25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
   26  *
   27  * $FreeBSD$
   28  *
   29  */
   30 
   31 /* 
   32  * This file includes definitions, structures, prototypes, and inlines that
   33  * should not be used outside of the actual implementation of UMA.
   34  */
   35 
   36 /* 
   37  * Here's a quick description of the relationship between the objects:
   38  *
   39  * Kegs contain lists of slabs which are stored in either the full bin, empty
   40  * bin, or partially allocated bin, to reduce fragmentation.  They also contain
   41  * the user supplied value for size, which is adjusted for alignment purposes
   42  * and rsize is the result of that.  The Keg also stores information for
   43  * managing a hash of page addresses that maps pages to uma_slab_t structures
   44  * for pages that don't have embedded uma_slab_t's.
   45  *  
   46  * The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may
   47  * be allocated off the page from a special slab zone.  The free list within a
   48  * slab is managed with a linked list of indexes, which are 8 bit values.  If
   49  * UMA_SLAB_SIZE is defined to be too large I will have to switch to 16bit
   50  * values.  Currently on alpha you can get 250 or so 32 byte items and on x86
   51  * you can get 250 or so 16byte items.  For item sizes that would yield more
   52  * than 10% memory waste we potentially allocate a separate uma_slab_t if this
   53  * will improve the number of items per slab that will fit.  
   54  *
   55  * Other potential space optimizations are storing the 8bit of linkage in space
   56  * wasted between items due to alignment problems.  This may yield a much better
   57  * memory footprint for certain sizes of objects.  Another alternative is to
   58  * increase the UMA_SLAB_SIZE, or allow for dynamic slab sizes.  I prefer
   59  * dynamic slab sizes because we could stick with 8 bit indexes and only use
   60  * large slab sizes for zones with a lot of waste per slab.  This may create
   61  * ineffeciencies in the vm subsystem due to fragmentation in the address space.
   62  *
   63  * The only really gross cases, with regards to memory waste, are for those
   64  * items that are just over half the page size.   You can get nearly 50% waste,
   65  * so you fall back to the memory footprint of the power of two allocator. I
   66  * have looked at memory allocation sizes on many of the machines available to
   67  * me, and there does not seem to be an abundance of allocations at this range
   68  * so at this time it may not make sense to optimize for it.  This can, of 
   69  * course, be solved with dynamic slab sizes.
   70  *
   71  * Kegs may serve multiple Zones but by far most of the time they only serve
   72  * one.  When a Zone is created, a Keg is allocated and setup for it.  While
   73  * the backing Keg stores slabs, the Zone caches Buckets of items allocated
   74  * from the slabs.  Each Zone is equipped with an init/fini and ctor/dtor
   75  * pair, as well as with its own set of small per-CPU caches, layered above
   76  * the Zone's general Bucket cache.
   77  *
   78  * The PCPU caches are protected by critical sections, and may be accessed
   79  * safely only from their associated CPU, while the Zones backed by the same
   80  * Keg all share a common Keg lock (to coalesce contention on the backing
   81  * slabs).  The backing Keg typically only serves one Zone but in the case of
   82  * multiple Zones, one of the Zones is considered the Master Zone and all
   83  * Zone-related stats from the Keg are done in the Master Zone.  For an
   84  * example of a Multi-Zone setup, refer to the Mbuf allocation code.
   85  */
   86 
   87 /*
   88  *      This is the representation for normal (Non OFFPAGE slab)
   89  *
   90  *      i == item
   91  *      s == slab pointer
   92  *
   93  *      <----------------  Page (UMA_SLAB_SIZE) ------------------>
   94  *      ___________________________________________________________
   95  *     | _  _  _  _  _  _  _  _  _  _  _  _  _  _  _   ___________ |
   96  *     ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header||
   97  *     ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________|| 
   98  *     |___________________________________________________________|
   99  *
  100  *
  101  *      This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE.
  102  *
  103  *      ___________________________________________________________
  104  *     | _  _  _  _  _  _  _  _  _  _  _  _  _  _  _  _  _  _  _   |
  105  *     ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i|  |
  106  *     ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_|  |
  107  *     |___________________________________________________________|
  108  *       ___________    ^
  109  *      |slab header|   |
  110  *      |___________|---*
  111  *
  112  */
  113 
  114 #ifndef VM_UMA_INT_H
  115 #define VM_UMA_INT_H
  116 
  117 #define UMA_SLAB_SIZE   PAGE_SIZE       /* How big are our slabs? */
  118 #define UMA_SLAB_MASK   (PAGE_SIZE - 1) /* Mask to get back to the page */
  119 #define UMA_SLAB_SHIFT  PAGE_SHIFT      /* Number of bits PAGE_MASK */
  120 
  121 #define UMA_BOOT_PAGES          48      /* Pages allocated for startup */
  122 
  123 /* Max waste before going to off page slab management */
  124 #define UMA_MAX_WASTE   (UMA_SLAB_SIZE / 10)
  125 
  126 /*
  127  * I doubt there will be many cases where this is exceeded. This is the initial
  128  * size of the hash table for uma_slabs that are managed off page. This hash
  129  * does expand by powers of two.  Currently it doesn't get smaller.
  130  */
  131 #define UMA_HASH_SIZE_INIT      32              
  132 
  133 /* 
  134  * I should investigate other hashing algorithms.  This should yield a low
  135  * number of collisions if the pages are relatively contiguous.
  136  *
  137  * This is the same algorithm that most processor caches use.
  138  *
  139  * I'm shifting and masking instead of % because it should be faster.
  140  */
  141 
  142 #define UMA_HASH(h, s) ((((unsigned long)s) >> UMA_SLAB_SHIFT) &        \
  143     (h)->uh_hashmask)
  144 
  145 #define UMA_HASH_INSERT(h, s, mem)                                      \
  146                 SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h),      \
  147                     (mem))], (s), us_hlink);
  148 #define UMA_HASH_REMOVE(h, s, mem)                                      \
  149                 SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h),           \
  150                     (mem))], (s), uma_slab, us_hlink);
  151 
  152 /* Hash table for freed address -> slab translation */
  153 
  154 SLIST_HEAD(slabhead, uma_slab);
  155 
  156 struct uma_hash {
  157         struct slabhead *uh_slab_hash;  /* Hash table for slabs */
  158         int             uh_hashsize;    /* Current size of the hash table */
  159         int             uh_hashmask;    /* Mask used during hashing */
  160 };
  161 
  162 /*
  163  * Structures for per cpu queues.
  164  */
  165 
  166 struct uma_bucket {
  167         LIST_ENTRY(uma_bucket)  ub_link;        /* Link into the zone */
  168         int16_t ub_cnt;                         /* Count of free items. */
  169         int16_t ub_entries;                     /* Max items. */
  170         void    *ub_bucket[];                   /* actual allocation storage */
  171 };
  172 
  173 typedef struct uma_bucket * uma_bucket_t;
  174 
  175 struct uma_cache {
  176         uma_bucket_t    uc_freebucket;  /* Bucket we're freeing to */
  177         uma_bucket_t    uc_allocbucket; /* Bucket to allocate from */
  178         u_int64_t       uc_allocs;      /* Count of allocations */
  179         u_int64_t       uc_frees;       /* Count of frees */
  180 };
  181 
  182 typedef struct uma_cache * uma_cache_t;
  183 
  184 /*
  185  * Keg management structure
  186  *
  187  * TODO: Optimize for cache line size
  188  *
  189  */
  190 struct uma_keg {
  191         LIST_ENTRY(uma_keg)     uk_link;        /* List of all kegs */
  192 
  193         struct mtx      uk_lock;        /* Lock for the keg */
  194         struct uma_hash uk_hash;
  195 
  196         LIST_HEAD(,uma_zone)    uk_zones;       /* Keg's zones */
  197         LIST_HEAD(,uma_slab)    uk_part_slab;   /* partially allocated slabs */
  198         LIST_HEAD(,uma_slab)    uk_free_slab;   /* empty slab list */
  199         LIST_HEAD(,uma_slab)    uk_full_slab;   /* full slabs */
  200 
  201         u_int32_t       uk_recurse;     /* Allocation recursion count */
  202         u_int32_t       uk_align;       /* Alignment mask */
  203         u_int32_t       uk_pages;       /* Total page count */
  204         u_int32_t       uk_free;        /* Count of items free in slabs */
  205         u_int32_t       uk_size;        /* Requested size of each item */
  206         u_int32_t       uk_rsize;       /* Real size of each item */
  207         u_int32_t       uk_maxpages;    /* Maximum number of pages to alloc */
  208 
  209         uma_init        uk_init;        /* Keg's init routine */
  210         uma_fini        uk_fini;        /* Keg's fini routine */
  211         uma_alloc       uk_allocf;      /* Allocation function */
  212         uma_free        uk_freef;       /* Free routine */
  213 
  214         struct vm_object        *uk_obj;        /* Zone specific object */
  215         vm_offset_t     uk_kva;         /* Base kva for zones with objs */
  216         uma_zone_t      uk_slabzone;    /* Slab zone backing us, if OFFPAGE */
  217 
  218         u_int16_t       uk_pgoff;       /* Offset to uma_slab struct */
  219         u_int16_t       uk_ppera;       /* pages per allocation from backend */
  220         u_int16_t       uk_ipers;       /* Items per slab */
  221         u_int32_t       uk_flags;       /* Internal flags */
  222 };
  223 
  224 /* Simpler reference to uma_keg for internal use. */
  225 typedef struct uma_keg * uma_keg_t;
  226 
  227 /* Page management structure */
  228 
  229 /* Sorry for the union, but space efficiency is important */
  230 struct uma_slab_head {
  231         uma_keg_t       us_keg;                 /* Keg we live in */
  232         union {
  233                 LIST_ENTRY(uma_slab)    _us_link;       /* slabs in zone */
  234                 unsigned long   _us_size;       /* Size of allocation */
  235         } us_type;
  236         SLIST_ENTRY(uma_slab)   us_hlink;       /* Link for hash table */
  237         u_int8_t        *us_data;               /* First item */
  238         u_int8_t        us_flags;               /* Page flags see uma.h */
  239         u_int8_t        us_freecount;   /* How many are free? */
  240         u_int8_t        us_firstfree;   /* First free item index */
  241 };
  242 
  243 /* The standard slab structure */
  244 struct uma_slab {
  245         struct uma_slab_head    us_head;        /* slab header data */
  246         struct {
  247                 u_int8_t        us_item;
  248         } us_freelist[1];                       /* actual number bigger */
  249 };
  250 
  251 /*
  252  * The slab structure for UMA_ZONE_REFCNT zones for whose items we
  253  * maintain reference counters in the slab for.
  254  */
  255 struct uma_slab_refcnt {
  256         struct uma_slab_head    us_head;        /* slab header data */
  257         struct {
  258                 u_int8_t        us_item;
  259                 u_int32_t       us_refcnt;
  260         } us_freelist[1];                       /* actual number bigger */
  261 };
  262 
  263 #define us_keg          us_head.us_keg
  264 #define us_link         us_head.us_type._us_link
  265 #define us_size         us_head.us_type._us_size
  266 #define us_hlink        us_head.us_hlink
  267 #define us_data         us_head.us_data
  268 #define us_flags        us_head.us_flags
  269 #define us_freecount    us_head.us_freecount
  270 #define us_firstfree    us_head.us_firstfree
  271 
  272 typedef struct uma_slab * uma_slab_t;
  273 typedef struct uma_slab_refcnt * uma_slabrefcnt_t;
  274 
  275 /*
  276  * These give us the size of one free item reference within our corresponding
  277  * uma_slab structures, so that our calculations during zone setup are correct
  278  * regardless of what the compiler decides to do with padding the structure
  279  * arrays within uma_slab.
  280  */
  281 #define UMA_FRITM_SZ    (sizeof(struct uma_slab) - sizeof(struct uma_slab_head))
  282 #define UMA_FRITMREF_SZ (sizeof(struct uma_slab_refcnt) -       \
  283     sizeof(struct uma_slab_head))
  284 
  285 /*
  286  * Zone management structure 
  287  *
  288  * TODO: Optimize for cache line size
  289  *
  290  */
  291 struct uma_zone {
  292         char            *uz_name;       /* Text name of the zone */
  293         struct mtx      *uz_lock;       /* Lock for the zone (keg's lock) */
  294         uma_keg_t       uz_keg;         /* Our underlying Keg */
  295 
  296         LIST_ENTRY(uma_zone)    uz_link;        /* List of all zones in keg */
  297         LIST_HEAD(,uma_bucket)  uz_full_bucket; /* full buckets */
  298         LIST_HEAD(,uma_bucket)  uz_free_bucket; /* Buckets for frees */
  299 
  300         uma_ctor        uz_ctor;        /* Constructor for each allocation */
  301         uma_dtor        uz_dtor;        /* Destructor */
  302         uma_init        uz_init;        /* Initializer for each item */
  303         uma_fini        uz_fini;        /* Discards memory */
  304 
  305         u_int64_t       uz_allocs;      /* Total number of allocations */
  306         u_int64_t       uz_frees;       /* Total number of frees */
  307         u_int64_t       uz_fails;       /* Total number of alloc failures */
  308         uint16_t        uz_fills;       /* Outstanding bucket fills */
  309         uint16_t        uz_count;       /* Highest value ub_ptr can have */
  310 
  311         /*
  312          * This HAS to be the last item because we adjust the zone size
  313          * based on NCPU and then allocate the space for the zones.
  314          */
  315         struct uma_cache        uz_cpu[1];      /* Per cpu caches */
  316 };
  317 
  318 /*
  319  * These flags must not overlap with the UMA_ZONE flags specified in uma.h.
  320  */
  321 #define UMA_ZFLAG_PRIVALLOC     0x10000000      /* Use uz_allocf. */
  322 #define UMA_ZFLAG_INTERNAL      0x20000000      /* No offpage no PCPU. */
  323 #define UMA_ZFLAG_FULL          0x40000000      /* Reached uz_maxpages */
  324 #define UMA_ZFLAG_CACHEONLY     0x80000000      /* Don't ask VM for buckets. */
  325 
  326 #ifdef _KERNEL
  327 /* Internal prototypes */
  328 static __inline uma_slab_t hash_sfind(struct uma_hash *hash, u_int8_t *data);
  329 void *uma_large_malloc(int size, int wait);
  330 void uma_large_free(uma_slab_t slab);
  331 
  332 /* Lock Macros */
  333 
  334 #define ZONE_LOCK_INIT(z, lc)                                   \
  335         do {                                                    \
  336                 if ((lc))                                       \
  337                         mtx_init((z)->uz_lock, (z)->uz_name,    \
  338                             (z)->uz_name, MTX_DEF | MTX_DUPOK); \
  339                 else                                            \
  340                         mtx_init((z)->uz_lock, (z)->uz_name,    \
  341                             "UMA zone", MTX_DEF | MTX_DUPOK);   \
  342         } while (0)
  343             
  344 #define ZONE_LOCK_FINI(z)       mtx_destroy((z)->uz_lock)
  345 #define ZONE_LOCK(z)    mtx_lock((z)->uz_lock)
  346 #define ZONE_UNLOCK(z)  mtx_unlock((z)->uz_lock)
  347 
  348 /*
  349  * Find a slab within a hash table.  This is used for OFFPAGE zones to lookup
  350  * the slab structure.
  351  *
  352  * Arguments:
  353  *      hash  The hash table to search.
  354  *      data  The base page of the item.
  355  *
  356  * Returns:
  357  *      A pointer to a slab if successful, else NULL.
  358  */
  359 static __inline uma_slab_t
  360 hash_sfind(struct uma_hash *hash, u_int8_t *data)
  361 {
  362         uma_slab_t slab;
  363         int hval;
  364 
  365         hval = UMA_HASH(hash, data);
  366 
  367         SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) {
  368                 if ((u_int8_t *)slab->us_data == data)
  369                         return (slab);
  370         }
  371         return (NULL);
  372 }
  373 
  374 static __inline uma_slab_t
  375 vtoslab(vm_offset_t va)
  376 {
  377         vm_page_t p;
  378         uma_slab_t slab;
  379 
  380         p = PHYS_TO_VM_PAGE(pmap_kextract(va));
  381         slab = (uma_slab_t )p->object;
  382 
  383         if (p->flags & PG_SLAB)
  384                 return (slab);
  385         else
  386                 return (NULL);
  387 }
  388 
  389 static __inline void
  390 vsetslab(vm_offset_t va, uma_slab_t slab)
  391 {
  392         vm_page_t p;
  393 
  394         p = PHYS_TO_VM_PAGE(pmap_kextract(va));
  395         p->object = (vm_object_t)slab;
  396         p->flags |= PG_SLAB;
  397 }
  398 
  399 static __inline void
  400 vsetobj(vm_offset_t va, vm_object_t obj)
  401 {
  402         vm_page_t p;
  403 
  404         p = PHYS_TO_VM_PAGE(pmap_kextract(va));
  405         p->object = obj;
  406         p->flags &= ~PG_SLAB;
  407 }
  408 
  409 /*
  410  * The following two functions may be defined by architecture specific code
  411  * if they can provide more effecient allocation functions.  This is useful
  412  * for using direct mapped addresses.
  413  */
  414 void *uma_small_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait);
  415 void uma_small_free(void *mem, int size, u_int8_t flags);
  416 #endif /* _KERNEL */
  417 
  418 #endif /* VM_UMA_INT_H */

Cache object: e8b86739b702f85ff8d9f6089aecf9c7


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.